[e90335e36m3]

Originally Posted by techwhiz1

Beyond L2 you are allowed to remove your hands from the wheel.

To support that function, you must provide graceful fail. If you do not have a secondary sensor you cannot positively locate in the lane if the primary sensor fails. Teslas are blind on sensor failure. For L3 and above you cannot be blind on a single point of failure.

That is the problem with Tesla ADAS; a single camera failure immediately renders some function unusable without graceful failure.

L3 requires redundancy that Tesla only has on their processor clusters. Elon had engineering remove all secondary sensors several years ago.

This is why Cruise, BMW, Mercedes and Ford can all offer hands free driving of some sort and Tesla cannot.

Also with a camera only system; fog, rain and other conditions can render the systems non functional.

I'm not saying it's currently or near term feasible, I'm saying that it may and likely will be feasible in the future, so using absolutes about its impossibility is not really prudent because they are proven wrong so often. I'm not one of these brilliant people but I work with aerospace and defence engineers and the stuff they create really boggles the mind. I'm put a wager 1000 to 1 that fully autonomous driving vehicles will be a reality, and what's going to slow them down will regulatory rather than technological in nature.

[GoldfishTX]

Originally Posted by bavarianride

GM did announce $0.7B 40000 L2 public charger network.

The 200k x $7.5k = $1.5B was what Fed earmarked to pop up each EV maker including Tesla.

Tesla used that $1.5B (that could stay in its coffer after credit expired) to expand its business including supercharger infrastructure.

And Tesla did drop prices when fed credit phased out.

Yes, they continued to reduce prices after the tax credit expired. They had already been going down for year leading up to reaching the 200k cap. They also raised and lowered prices many times after that point in time by similar amounts. Tesla also offered to pay customers the difference between the expected and actual tax credit out of pocket when it phased out. This wasn't some 7.5k/vehicle money printing machine like you imply. Once again, this tax credit that went to the buyers did not fund the supercharger network.

[bavarianride]

Originally Posted by pikcachu

Tesla supercharger network costed way more than 1.5B

1.5B is like a drop in the bucket

The numbers I read were around $0.25m per supercharger stations of various number of ports. So the current 1200 superchargers costed around $3B.

$1.5B US taxpayer money can fund half of those, so I am OK with 50% of supercharger station ports opened to the public.

[pikcachu]

Originally Posted by bavarianride

The numbers I read were around $0.25m per supercharger stations of various number of ports. So the current 1200 superchargers costed around $3B.

$1.5B US taxpayer money can fund half of those, so I am OK with 50% of supercharger station ports opened to the public.

No. Each supercharger station costs several Millions.

If it was just 3B why do you think no other manufacturer has its own ?

[antzcrashing]

Originally Posted by pikcachu

No. Each supercharger station costs several Millions.

If it was just 3B why do you think no other manufacturer has its own ?

I have seen estimates of 100k - 250k. Source for 1million+?

To answer your rhetorical question, the big manufacturers didn't believe in EVs like Tesla did, nor did they think range anxiety needed an engineering infrastructure solution.

[bavarianride]

Originally Posted by pikcachu

No. Each supercharger station costs several Millions.

If it was just 3B why do you think no other manufacturer has its own ?

Several millions for each station sound a bit high. UBS had an analysis that estimated $250k per station a few years ago(2017-2018?), I haven't seen any number disclosed in latest Tesla annual report.

Can u provide a link/source to the the cost of several $m per Tesla supercharger station?

VW does put $2B Electrify America for 800 stations and 3500 chargers(reached 635 stations and 2700 chargers in 7/2021).

BMW, Ford, Hyundai, KIA, Mercedes and VW do join force in Europe to push IONITY, which recently secured €0.7B for extra 5000 350kW charging ports in Europe by 2025.

[techwhiz1]

Originally Posted by bavarianride

That's why I own Tesla's best product, which is not its cars.

Now the PowerWall is a solid product.

I got Tesla Solar before they started doing all the leasing stuff.

[nosnoop]

Originally Posted by antzcrashing

I agree with you that camera-based autonomous tech alone may never be able to hit Level 5 in all weather and driving situations, but that does not prevent tesla from adding a Lidar system and supplementing it with the camera-based software that they have developed. Nor does it prevent them from doing a complete 180 and switching to Lidar. Lidar is not rocket science when it comes to assisted braking, etc.

Tesla will never use Lidar as long as Elon is in charge.

[giphlag]

Originally Posted by techwhiz1

As for ADAS? Surely you are joking. Tesla is L2 with hands on as a requirement even with optional FSD. Tesla uses a camera only system with no redundancy in the cameras which guarantees it will never be FSD.

That is a pretty bold statement to make and I wonder where you get this information.
Teslas have full 360 degree camera vision. If one camera fails or its view is obscured the others take over in combination with their predictive modeling. The car is programmed to come to a graceful and controlled stop or put the burden of driving back on the rider.

I personally find the doubling down on full Machine Vision at Tesla fascinating. I'm impressed by their technology stack, neural network and latest Dojo chip design and system. Like someone else said. Don't count them out completely, just because they aren't doing it like everyone else.

[BimmerBahn]

Originally Posted by giphlag

That is a pretty bold statement to make and I wonder where you get this information.
Teslas have full 360 degree camera vision. If one camera fails or its view is obscured the others take over in combination with their predictive modeling. The car is programmed to come to a graceful and controlled stop or put the burden of driving back on the rider.

I personally find the doubling down on full Machine Vision at Tesla fascinating. I'm impressed by their technology stack, neural network and latest Dojo chip design and system. Like someone else said. Don't count them out completely, just because they aren't doing it like everyone else.

full (L4+) autonomy driving has a - long - way to go.... I don't necessarily think vision-only was the right decision... I think it had to do more with cost savings than anything else.... but time will tell. I expect the camera system to iterate along with the FSD computer. I do agree that Dojo is potentially a game changer for them.... it's a custom ASIC that is very powerful - needed to do very large scale DNN training. They've still got a long road ahead.

I read recently where a semi w/ full autonomous driving successfully navigated ~80 miles without any human intervention... and no one in the cab, either. Apparently there was a spotter running ahead and police support as well. But that IS a big deal the fact it was able to transit on both freeway and local roads 100% by itself.

https://www.fox10phoenix.com/news/au...st-in-arizona1

[bavarianride]

Originally Posted by giphlag

That is a pretty bold statement to make and I wonder where you get this information.
Teslas have full 360 degree camera vision. If one camera fails or its view is obscured the others take over in combination with their predictive modeling. The car is programmed to come to a graceful and controlled stop or put the burden of driving back on the rider.

I personally find the doubling down on full Machine Vision at Tesla fascinating. I'm impressed by their technology stack, neural network and latest Dojo chip design and system. Like someone else said. Don't count them out completely, just because they aren't doing it like everyone else.

Tesla actually triples down on computer vision by not using lidar, and getting rid of radar.

I think Techwiz's point is that FSD requires redundant fail-safe, and Tesla's camera-only without overlapping camera per angle(up to 8), nor overlapping camera + radar + lidar, does not have redundancy.

It appears J3016 level 1 to 5 by itself does not define redundancy nor safety, so technically Tesla can build and claim L1-L5 as they see fit.

My take is that J3016 should define safety requirements of level 1-5 perception systems, and IIHS/NHTSA should test and rate them per spec to rein in the wild wild west.

[bavarianride]

Originally Posted by BimmerBahn

full (L4+) autonomy driving has a - long - way to go.... I don't necessarily think vision-only was the right decision... I think it had to do more with cost savings than anything else.... but time will tell. I expect the camera system to iterate along with the FSD computer. I do agree that Dojo is potentially a game changer for them.... it's a custom ASIC that is very powerful - needed to do very large scale DNN training. They've still got a long road ahead.

I read recently where a semi w/ full autonomous driving successfully navigated ~80 miles without any human intervention... and no one in the cab, either. Apparently there was a spotter running ahead and police support as well. But that IS a big deal the fact it was able to transit on both freeway and local roads 100% by itself.

https://www.fox10phoenix.com/news/au...st-in-arizona1

What was the speed of that 80-mile test?

It looks like Dojo is a supercomputer chip that catches up with Nvidia and TPU offerings, but Dojo is still central training, not edge, and cannot be used on the car itself due to power constraints.

To get real time data acquisition + real time central training + re-infer at edge, at 70-100+mph, likely will take a long while. Obviously spotter + police assistance can help a lot.

[BimmerBahn]

Originally Posted by bavarianride

What was the speed of that 80-mile test?

It looks like Dojo is a supercomputer chip that catches up with Nvidia and TPU offerings, but Dojo is still central training, not edge, and cannot be used on the car itself due to power constraints.

To get real time data acquisition + real time central training + re-infer at edge, at 70-100+mph, likely will take a long while. Obviously spotter + police assistance can help a lot.

Not sure, but from the video, it looks like it was "highway speed", following speed limit and other road signs.

Agree on Dojo... I expect the Dojo tech to get scaled down for edge use in the FSD computer... a 400W ASIC package isn't realistic for a mobile edge where power/cooling is limited... I think current FSD computer is ~100W for the whole thing. It's still very powerful for the platform, but not enough for "FSD"

[techwhiz1]

Originally Posted by bavarianride

Tesla actually triples down on computer vision by not using lidar, and getting rid of radar.

I think Techwiz's point is that FSD requires redundant fail-safe, and Tesla's camera-only without overlapping camera per angle(up to 8), nor overlapping camera + radar + lidar, does not have redundancy.

It appears J3016 level 1 to 5 by itself does not define redundancy nor safety, so technically Tesla can build and claim L1-L5 as they see fit.

My take is that J3016 should define safety requirements of level 1-5 perception systems, and IIHS/NHTSA should test and rate them per spec to rein in the wild wild west.

You are correct that redundancy isn't defined, but by usage you can infer redundancy.

If your hands are off the wheel for prolonged periods then any period of blindness isn't acceptable.

if you overlay the L0-L5 usage on ISO26262 (ASIL A-D) redundancy in vision and processing is defined when injury or death can occur.

I would argue, that when a car is driving at L3 and above death or injury are imminent if you don't have redundancy.

The other obvious thing is with a vision only system inclement weather usage is almost out. Tesla insistance on using vision with no static maps for lane location means that they must always have a pretty clear view of everything.

I'm not the final say so, I know enough that eight cameras can provide limited redundancy but it is very limited and they need to be very strategically placed not to be obscured by the car itself

For comparison BMW, GM, Ford and Mercedes all use vision coupled with Radar, Lidar and ultrasound (PDC short range) and they all have some degree.of hands off and only driver awareness is required. Tesla isn't there yet. Mercedes has applied for L3 in Europe. BMW has indicated they can want L3 from US regulators.

MBZ uses Nvidia Drive platform and they have demonstrated FSD including merge, roundabout, pedestrians, etc. The video can be found as part of the Nvidia GTC in November.

I like, the rest of the industry find it highly doubtful that FSD from Tesla will reach the market with 8 Vision cameras and no other systems.

But we do digress. I'd much rather talk about how the M50 doesn't seem to be "smoked" by a M3P as some people had claimed by looking. at BMW specs.

[pikcachu]

Originally Posted by techwhiz1

You are correct that redundancy isn't defined, but by usage you can infer redundancy.

If your hands are off the wheel for prolonged periods then any period of blindness isn't acceptable.

if you overlay the L0-L5 usage on ISO26262 (ASIL A-D) redundancy in vision and processing is defined when injury or death can occur.

I would argue, that when a car is driving at L3 and above death or injury are imminent if you don't have redundancy.

The other obvious thing is with a vision only system inclement weather usage is almost out. Tesla insistance on using vision with no static maps for lane location means that they must always have a pretty clear view of everything.

I'm not the final say so, I know enough that eight cameras can provide limited redundancy but it is very limited and they need to be very strategically placed not to be obscured by the car itself

For comparison BMW, GM, Ford and Mercedes all use vision coupled with Radar, Lidar and ultrasound (PDC short range) and they all have some degree.of hands off and only driver awareness is required. Tesla isn't there yet. Mercedes has applied for L3 in Europe. BMW has indicated they can want L3 from US regulators.

MBZ uses Nvidia Drive platform and they have demonstrated FSD including merge, roundabout, pedestrians, etc. The video can be found as part of the Nvidia GTC in November.

I like, the rest of the industry find it highly doubtful that FSD from Tesla will reach the market with 8 Vision cameras and no other systems.

But we do digress. I'd much rather talk about how the M50 doesn't seem to be "smoked" by a M3P as some people had claimed by looking. at BMW specs.

Yeah it's pretty interesting how close they are.

Not quite sure why the specs are so out of wack with real life

I guess the downside is the extra weight which hurts braking and handling and the lower range

But you get the better build quality and it looks pretty good although that always quite subjective

I guess the price adds up too but performance wise seems like a toss

[pikcachu]

But if you want something you can take to the track the M3 xdrive is still the answer.

And a Plaid for stop light joy

[bavarianride]

Originally Posted by pikcachu

But if you want something you can take to the track the M3 xdrive is still the answer.

And a Plaid for stop light joy

Yeah if in doubt, just buy them all.

[techwhiz1]

Originally Posted by pikcachu

Yeah it's pretty interesting how close they are.

Not quite sure why the specs are so out of wack with real life

I guess the downside is the extra weight which hurts braking and handling and the lower range

But you get the better build quality and it looks pretty good although that always quite subjective

I guess the price adds up too but performance wise seems like a toss

So I'm not sure about the braking yet.
Did they have the 200kW Regen active?
Did they panic stop?

Since It's brake by wire, the setup is going to matter a lot.

As far as handling, BMW knows how to tune suspension for heavy cars. The M5 handles well and it's about 4400lbs.

The car looked good going around the track.

[NISFAN]

Originally Posted by techwhiz1

So I'm not sure about the braking yet.
Did they have the 200kW Regen active?
Did they panic stop?

Since It's brake by wire, the setup is going to matter a lot.

As far as handling, BMW knows how to tune suspension for heavy cars. The M5 handles well and it's about 4400lbs.

The car looked good going around the track.

Yes full emergency stop, on the ABS.

Beaten by a 20k mile Tesla with typical EV weathered brakes. Note how one of the brake pads is not even wiping the full rotor surface.

Well at least we know Newton's 2nd law is still intact.

[techwhiz1]

Originally Posted by NISFAN

Yes full emergency stop, on the ABS.

Beaten by a 20k mile Tesla with typical EV weathered brakes. Note how one of the brake pads is not even wiping the full rotor surface.

Well at least we know Newton's 2nd law is still intact.

I had to go back and look and yes the pad isn't making full contact along the full pad height.

I really want to know the BMW setup though.

[bavarianride]

Originally Posted by techwhiz1

I had to go back and look and yest the pad isn't making full contact along the full pad height.

I really want to know the BMW setup though.

Does it mean BMW can improve pad contact on rotor and reduce braking distance?

I am still surprised that 1000+lb only has a car length difference in braking distance 100mph to 0mph, the regenerative brakes through electric motors likely help a lot.

[NISFAN]

Originally Posted by bavarianride

Does it mean BMW can improve pad contact on rotor and reduce braking distance?

I am still surprised that 1000+lb only has a car length difference in braking distance 100mph to 0mph, the regenerative brakes through electric motors likely help a lot.

No, the 20k mile Tesla is the one in that test with suspect brakes…..and suspect 1/4 ET but that is another story.

Brake regeneration won't have any effect on a panic stop, the ABS will be holding friction brake force at tyre grip limit, with brake force reserves from the friction brakes.
The M50 is a heavy car and unlike the G82 where fatter tyres can be thrown at it to combat the weight, that's not a practical strategy on an EV. We can already see a significant drop in range due to the 20" wheel set option, wider rubber would make that even worse. You can't have your cake and eat it, as they say.

I'd like to see same test with M50 on standard 19" vs the optional 20". Would the 19" accelerate faster?

Other than avoiding a collision, I don't think it is detrimental to the BMW that it doesn't stop as quick as a TM3P, you aren't likely to track an M50, and if you were you just drop some R compound rubber on it. Matt mentioned issue of LC not engaging after another launch due to battery temps. I suspect that will become an issue on track, battery temp management that is. We know Tesla has revised battery cooling over the years and they still have problems completing an entire Nurburgring lap without power reduction due to batt temp.

Anyone know what LC does on the M50? Is it only in this mode that the power boost is active? AFAIK TM3P has no launch mode, select sport drive setting and it smashes out 3.3 0-60 all day long. As you wouldn't be choosing "chill mode" in a Model 3 it would be default sport.